In the energy context, part of the challenge is using new infrastructure like solar panels to meet new demands such as those of electric car drivers or heat-pump users. (Pictured: solar panels installed by Luxtram.) Photo: Luxtram

In the energy context, part of the challenge is using new infrastructure like solar panels to meet new demands such as those of electric car drivers or heat-pump users. (Pictured: solar panels installed by Luxtram.) Photo: Luxtram

If everyone started driving an electric car tomorrow, could the energy grid handle it? What about in 15 years? LIST researcher Jun Cao spoke to Delano about the energy transition and how artificial intelligence might play a central role.

The transition to sustainable energy is firmly underway, both in terms of technological ability and societal willpower. Greener versions of cars, heaters and generators are increasingly seen not only as a trend but an existential imperative.

Hardware and willpower are crucial, but the transition requires more—it requires the energy sector to ramp up its supply of electricity to meet demands that are expected to skyrocket.

Indeed, there were about 1.3m electric vehicles in Europe in 2019, according to data from Virta, a number that is on track to hit 14m in 2025. Five years after that, as many as 40m electric cars are foreseen, and by 2035 all new cars sold in Europe will be electric. Heat pumps, which transfer thermal energy indoors or outdoors for heating/cooling purposes, are another example of a burgeoning green technology: installing its predecessor, the gas heater, isn’t even permitted in Luxembourg anymore.

Further complicating energy grids are new sources of electricity generation, namely solar panels and wind turbines, whose outputs are highly variable as they depend (respectively) on sunlight and wind.

And finally, on top of all that, comes the digitalisation of energy systems. This translates into huge volumes of data that need to be parsed, understood and used, for example the readouts of the smart meters that are increasingly deployed .

During a recent event hosted by the Luxembourg Institute of Science and Technology (LIST), researcher Jun Cao catalogued these factors—new eco-friendly designs, new energy-generation technologies, exponentially larger volumes of data—before adding a fourth one: user behaviour. Average energy consumers are turning into prosumers, he said, or nonspecialists who are using professional-grade electronics. Individuals will be able to buy, sell or store energy, meaning that their decisions will affect supply and demand.

“Users will play an essential role in the future energy transition,” Cao commented, “from design to operation to consumption. With traditional model-based methods, it is hard to predict or model this user behaviour, so we need a new algorithm.”

Artificial intelligence

All of these factors require, Cao pointed out, a paradigm shift. His research at LIST is geared towards enabling this shift using artificial intelligence and machine learning.

“My dream of future energy systems,” he told Delano in an interview, “is for every household, building or renewable energy community in the future to have their own AI-powered energy management toolbox (I call it a ‘personal energy assistant’). Just like your TV box, it will learn your behaviour and manage the energy generation and consumption in a more intelligent, efficient and environmentally friendly way.”

Before such a dream can be realised, however, a few obstacles must be overcome. From a research standpoint, Cao said, data poses some problems: it isn’t always available for research purposes, the worlds of industry and academia don’t always collaborate (effectively), privacy questions need answering.

Another stumbling block surrounds the issue of trustworthiness. “AI in the energy field shouldn’t be a black box,” the researcher said. “It should be transparent, explainable, robust and safe.”

These challenges relate to the main risks of AI-powered or autonomous energy systems, which also involve data. “If there is data attack or the input data is manipulated,” Cao explained, “the decision made from the bad data will be wrong, which will threaten the operation of the energy system.” Detecting AI cyberattacks is its own research field, he added, which is tackling just this problem.

Timeline

In all, Cao estimated that his vision of an AI-powered energy grid isn’t exactly around the corner. “I would say there is a long journey to reach the destination,” he commented, citing the necessity for absolute certainty regarding safety issues.

“But,” he added, “we are starting to see changes on the user/community side, with more and more renewable energy (like rooftop solar panels) and electric vehicles. The AI-powered personal energy assistant will become reality.”